talking-head-analyzer
Warn
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local bash scripts (scripts/prepare_clips.sh and scripts/contact_sheet.sh) that use ffmpeg for video manipulation. Using shell scripts with file-based inputs presents a surface for potential argument injection if filenames are not sanitized by the calling environment.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted video transcripts and frame data to create new AI instructions. Malicious content within the ingested media could be carried over into the generated skill files. Evidence: 1. Ingestion points: Video files and transcripts (audio/*.wav) processed in SKILL.md. 2. Boundary markers: None identified to isolate transcript content. 3. Capability inventory: File writing (SKILL.md), media processing (ffmpeg), and directory creation. 4. Sanitization: No explicit validation or filtering of transcript content before template interpolation.
- [EXTERNAL_DOWNLOADS]: The skill instructions reference external tools such as yt-dlp for downloading content and mlx_whisper for processing audio, which involve external network requests and third-party software dependencies.
Audit Metadata